engleski

Post-processing of ALADIN forecasts using neighbourhood techniques​

It is important to understand that the model grid size is not the same as the model resolution. The second is sometimes referred to as the model effective resolution and is generally, at least, 5 times lower than the first. In addition, lowering the grid spacing leads to faster error growth and saturation on the smallest resolved scales. For kilometric grid sizes, error saturation can occur after only a couple of hours of integration. This means that model forecasts on those scales become uncertain very quickly. Therefore, all point predictions within that area (i.e., neighborhood) should be considered equally likely and the output of the model should be viewed as the spatial and (or) temporal function of that neighborhood. To alleviate before-mentioned difficulties, neighborhood methods were developed for: a) The use in the forecast verification as spatial verification methods where they generally share a common trait of relaxing the traditional requirement that forecast and observed events exactly match at the grid scale to account for observation and model uncertainties. b) To extend an EPS by increasing the number of its members and (or) to provide a way to calculate ensemble probabilities which better reflect the model’s true resolution. In this work, we apply neighbourhood method to a deterministic ALADIN forecast. Selected neighborhood contains both spatial and temporal dimension and its size varies with the forecast range to account for increasing forecast uncertainty. By using neighbourhoods, we can include probabilistic information to and reduce representativeness error of a deterministic forecast. We apply this technique to precipitation, temperature, and wind variables. The results show increased forecast accuracy for all variables, especially for min/max temperatures and it gives us an elegant way to account for double-penalty effect for precipitation. In addition, various forecast products based on the neighborhood approach will be presented.

ALADIN, neighborhood, ensemble, post-processing

nije evidentirano